thek0tyara/llama-cpp-turboquant
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions 4

Reformatted

Browse source
This commit is contained in:
Alejandro Saucedo 2020-08-22 13:01:16 +01:00
parent 25d5df21ce
commit c92425dd87
3 changed files with 28 additions and 23 deletions
Download patch file Download diff file Expand all files Collapse all files

16
src/OpMult.cpp
View file

@ -86,18 +86,18 @@ OpMult::record()
this->mAlgorithm->recordDispatch(1, 1, 1);
this->mTensorOutput->recordBufferMemoryBarrier(
vk::AccessFlagBits::eShaderWrite,
vk::AccessFlagBits::eTransferRead,
vk::PipelineStageFlagBits::eComputeShader,
vk::PipelineStageFlagBits::eTransfer);
vk::AccessFlagBits::eShaderWrite,
vk::AccessFlagBits::eTransferRead,
vk::PipelineStageFlagBits::eComputeShader,
vk::PipelineStageFlagBits::eTransfer);
this->mTensorOutputStaging->recordCopyFrom(this->mTensorOutput);
this->mTensorOutput->recordBufferMemoryBarrier(
vk::AccessFlagBits::eTransferWrite,
vk::AccessFlagBits::eHostRead,
vk::PipelineStageFlagBits::eTransfer,
vk::PipelineStageFlagBits::eHost);
vk::AccessFlagBits::eTransferWrite,
vk::AccessFlagBits::eHostRead,
vk::PipelineStageFlagBits::eTransfer,
vk::PipelineStageFlagBits::eHost);
}
void

27
src/Tensor.cpp
View file

@ -136,7 +136,12 @@ Tensor::recordCopyFrom(std::shared_ptr<Tensor> copyFromTensor)
this->mData = copyFromTensor->mData;
}
void Tensor::recordBufferMemoryBarrier(vk::AccessFlagBits srcAccessMask, vk::AccessFlagBits dstAccessMask, vk::PipelineStageFlagBits srcStageMask, vk::PipelineStageFlagBits dstStageMask) {
void
Tensor::recordBufferMemoryBarrier(vk::AccessFlagBits srcAccessMask,
vk::AccessFlagBits dstAccessMask,
vk::PipelineStageFlagBits srcStageMask,
vk::PipelineStageFlagBits dstStageMask)
{
SPDLOG_DEBUG("Kompute Tensor recording buffer memory barrier");
vk::DeviceSize bufferSize = this->memorySize();
@ -146,19 +151,15 @@ void Tensor::recordBufferMemoryBarrier(vk::AccessFlagBits srcAccessMask, vk::Acc
bufferMemoryBarrier.size = bufferSize;
bufferMemoryBarrier.srcAccessMask = srcAccessMask;
bufferMemoryBarrier.dstAccessMask = dstAccessMask;
bufferMemoryBarrier.srcQueueFamilyIndex =
VK_QUEUE_FAMILY_IGNORED;
bufferMemoryBarrier.dstQueueFamilyIndex =
VK_QUEUE_FAMILY_IGNORED;
this->mCommandBuffer->pipelineBarrier(
srcStageMask,
dstStageMask,
vk::DependencyFlags(),
nullptr,
bufferMemoryBarrier,
nullptr);
bufferMemoryBarrier.srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED;
bufferMemoryBarrier.dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED;
this->mCommandBuffer->pipelineBarrier(srcStageMask,
dstStageMask,
vk::DependencyFlags(),
nullptr,
bufferMemoryBarrier,
nullptr);
}
// TODO: Explore if this function should be here or expose buffer

8
src/Tensor.hpp
View file

@ -51,8 +51,12 @@ class Tensor
// Record functions
void recordCopyFrom(std::shared_ptr<Tensor> copyFromTensor);
// TODO: Explore simplifying by infering pipeline stage flag bits from access flag bits (as seems to be superset)
void recordBufferMemoryBarrier(vk::AccessFlagBits srcAccessMask, vk::AccessFlagBits dstAccessMask, vk::PipelineStageFlagBits srcStageMask, vk::PipelineStageFlagBits dstStageMask);
// TODO: Explore simplifying by infering pipeline stage flag bits from
// access flag bits (as seems to be superset)
void recordBufferMemoryBarrier(vk::AccessFlagBits srcAccessMask,
vk::AccessFlagBits dstAccessMask,
vk::PipelineStageFlagBits srcStageMask,
vk::PipelineStageFlagBits dstStageMask);
// Util functions
vk::DescriptorBufferInfo constructDescriptorBufferInfo();